Harvard, apparently, is using virtual faculty on one of its courses — and charging $699 for the privilege!! Per Sarah Kessler, writing in The Star:
I’ve just started a video call with a general partner at a venture capital firm. “Whenever you’re ready, please start your pitch,” he says.
For a reporting experiment, I am playing the role of a hopelessly delusional tech founder. As I outline my plan to build an “Uber for bananas,” his calm, polite smile doesn’t move. Neither do the small shadows cast by the drape of his shirt. His shoulders hover back and forth just slightly – and a little too consistently. Because while it looks like I’m talking with Jeff Bussgang, the co-founder of Flybridge Capital and a senior lecturer at Harvard Business School, I’m actually talking with his AI-generated video avatar.
Well now.
If only I had built a virtual version of me, trained on all my writing and on extensive psychological profiling of me.1 Wait, I have! It’s read everything - even my Substacks. And it’s watched me lecture too - those are all filmed these days, for students to watch back in their own time. I’ve discussed Ken-bot here a while ago - he’s the testbed for my wider research project on building AI personas. After all, in theory I know me better than anyone else.
He’s evolved since. Today’s Kenbot is agentic - he can go online and search out any info he wants, and very soon he’ll be creating his own Powerpoints too, in my style to boot!2 He marks papers diligently, cohering with your university’s grading framework and offering constructive student feedback. In this, he’s doing for Higher Education what my Project Kennan is doing for think tanks.
So, could he take my courses? You bet! Prof Ken-bot, what are your reflections 4 years on from, ‘I, Warbot’? Has your argument held up?3
What does all this mean? Am I out of a job? I hope not, as I rather enjoy it. But why wouldn’t a university want an AI version of the prof that can guide students through online courses, like some over-educated Clippy? Or that can solve teaching bottlenecks in high volume short courses, where you might not otherwise see real faculty? Or that can assist new faculty teaching assistants who are less familiar with the modules they’re working on, sometimes at short notice?
Anyway, for me it’s a fun, if narcissistic side-quest to my bid to create synthetic personas and explore their ‘machine-psychology’ a bit more - just as Ken-bot said.
Here’s me versus Vance, on dark triad metrics, a subset of the much larger battery of tests I did. The main difference, according to the ensemble of LLM shrinks that weigh the data, he’s way more Machiavellian than me (and I’m actually way less Machiavellian than I think I am).
I particularly like that Ken-bot autonomously adds his own speaking notes to his ppts, as here…
(Note, there is a video version of Ken-bot too, but I’m not 100% happy with that yet…. Here’s the transcript:
